Cisco Systems 1000 Series Manual De Usuario

When packets transit the MLP transmit path, they are subject to two separate stages of queuing. The 
first stage is at the MLP bundle interface, where QoS may be applied, and the second one is at the 
MLP member-link interface. At the MLP bundle interface, the packets are processed according to 
the applied QoS policy. Packets classified as priority are given preferential treatment over 
nonpriority traffic.

For the priority classification to be honored at the MLP member-link interface, the bundle must have 
ppp multilink interleave enabled. Interleaving allows a packet to be queued to a separate priority 
queue at the member-link. If interleaving is not enabled on the bundle, the priority packet is placed 
in the member link session default queue and the knowledge that it is a priority packet will be lost. 
This is especially important if there are other PPP or MLP sessions sharing the same physical 
interface or subinterface. Without interleaving, priority traffic on the other sessions are given 
preferential treatment over the MLP priority packets that were reclassified as nonpriority packets at 
the MLP member-link queuing stage. For additional information on interleaving, see the 

On the Cisco ASR 1000 Series Aggregation Services Routers, Multilink PPP adjusts the packet length 
presented for shaping and policing to include the additional Layer 2 overhead added by Multilink PPP. 
For MLP over Serial, overhead accounting includes the MLP and PPP Layer 2 overhead. For Broadband 
MLPs such as MLPoE, MLPoEoVLAN, MLPoEoQinQ, MLPoEoA, MLPoA, and MLPoLNS, overhead 
accounting includes the MLP, PPP, Ethernet, ATM, and L2TP (LNS) Layer 2 overhead. If the output 
interface is ATM, such as the MLPoA or MLPoEoA, the Cisco ASR 1000 Series Aggregation Services 
Routers also take into account the ATM Cell overhead for the shaper. The ATM Cell overhead is not 
accounted for policing.

Shaping and policing overhead accounting does not include the additional overheads added by a SPA 
such as, Ethernet CRC, preamble, IPG, serial interface CRC, start of packet (SOP) delimiter, end of 
packet (EOP) delimiter, and serial-bit stuffing (the only exception being the ATM Cell overhead for the 
shaper referred to earlier). The overhead added by a SPA can be included in the shaper using the QoS 
shape accounting user-defined option.

If you do not define a QoS shaper for the multilink bundle interface, a default shaper is applied to the 
bundle based on the aggregate bandwidth of all the links that make up the multilink bundle. The 
information contained in this section applies to both the default shaper and a QoS user-defined shaper, 
which the user may explicitly configure and apply to a multilink bundle.

The priority packets that are interleaved are sent PPP encapsulated and the MLP Layer 2 overhead is not 
included because MLP encapsulation is not included in these packets. During overhead accounting for 
link fragmentation, overhead accounting calculations are performed prior to the actual link 
fragmentation and link selection for Multilink PPP load balancing. 

If all the member links in the corresponding multilink bundle use the same fragment size, the number of 
fragments are calculated and the overhead is adjusted to include the additional per-fragmentation Layer 
2 header overhead for the shaper and policer. If one or more links in the bundle use different fragment 
sizes, the number of fragments cannot be calculated with 100 percent accuracy because link selection 
for load balancing and fragment size is not known until QoS processing is completed at the bundle level 
(after shaping and policing). For links with unequal fragment size, a best effort attempt is made using 
the largest link fragment size on the bundle. By using the largest fragment size, MLP avoids 
undersubscribing the member-link interfaces. If the links become oversubscribed, MLP will 
backpressure the bundle to avoid sustained oversubscription of the member links.

In Cisco IOS XE Release 3.4S on the Cisco ASR 1000 Series Aggregation Services Routers, support for 
shaping and policing overhead accounting was added for Broadband Multilink PPP. In addition, support 
was added for the Shape User-Defined Overhead Accounting feature using the following QoS command: